摘要
背景及目的：肺癌是全球主要癌症死因。目前並無肺癌篩檢之有效策略。以胸部電腦斷層作為肺癌篩檢有許多爭論，目前有兩個大型隨機分派臨床試驗進行中（NELSON 及 NLST），但短期內不會有結果。故本研究目的在於（1）研究不同年齡及世代下之肺癌發生率（2）相應於不同篩檢工具下之肺癌平均滯留時間（3）配合上述（1）（2），發展依決策分析模式以作為肺癌篩檢之經濟評估。 方法：我們以決策分析軟體TreeAgePro建構一個五階段馬可夫模式以便模擬肺癌篩檢。主要評估點為有效性與成本效用，其中有效性以預期人年差異來表示，而成本效用以增加成本效用比來表示，其中效用以品質校正人年（QALY）來表示。另外，我們建立五階段馬可夫模式中所需之重要參數（發生率、平均滯留時間、及治療順從性），而其他所需之參數則以西方文獻為主。我們以貝氏年齡期間世代模式分析癌症登記資料來估計發生率。對相應於電腦斷層的平均滯留時間，我們用貝氏三階段馬可夫模式分析文獻報告來估計。至於相應於胸部愛克斯光的平均滯留時間，我們則用貝氏三階段馬可夫模式分析一伺機性篩檢族群來估計。我們另以回歸及描述性統計來分析醫院癌症登記資料以便估計治療順從性。在完成模式建構與參數估計後，我們比較了模式估計值與文獻報告來做模式驗證。比較的項目包含不同期別之個案發現率，不同期別之個案存活率，及不同期別之治療費用。最後我們以上述之模式與參數進行不同肺癌篩檢情境下之模擬。主要的模擬情境是參考NELSON試驗（61歲高危險性吸煙者進行三次電腦斷層篩檢或只是觀察），但我們也模擬了其他情境。 結果：我們發現台灣肺癌發生率就男性而言在公元2000年後已稍微下降，但就女性而言仍在稍微上升。在1996年，61歲台灣男性之肺癌發生率（1/100000人年）為102 （95%信賴區間：80∼127），而女性則為31（22∼44）。到了2002年，相對應的值是123(99~147)及58（45∼75）。我們估計肺癌相應於電腦斷層的平均滯留時間是2.06年（95%信賴區間：0.42∼3.83），而相應於胸部愛克斯光的平均滯留時間是5.51個月（95%信賴區間：4.04∼7.12）。我們估計晚期肺癌的治療順從性為40%。在模式驗證方面，我們發現模式估計值與文獻報告尚稱接近。在不同期別之個案發現率上，我們預估篩檢可在受檢者中發現1.1％早期肺癌及1.4％肺癌，此與文獻相近（1.3％ ＆ 1.5％）。在不同期別之個案存活率上，我們預估早期與晚期肺癌之五年存活率分別為62％∼77％ ＆ 0∼1％，此亦與文獻相近（54∼73％ ＆ 1∼7％）。至於不同期別之治療費用（美金，USD），我們預估早期與晚期肺癌分別為25183 ＆ 22372，此亦與文獻相近（25050 ＆ 20691）。最後，在我們的主要模擬情境（NELSON）上，我們預估預期人年差異為0.018 年（95％信賴區間-0.04∼0.07），而增加成本效用比（USD/QALY）為125171 （社會觀點）（如以付費者觀點則為 237145）。若考慮不同情境，如密集篩檢，每年電腦斷層篩檢相較於每三年一次可減少肺癌死亡率（2.1％ vs 2.4％）但不合成本效用。相較於愛克斯光篩檢，電腦斷層篩檢的增加成本效用比（USD/QALY）為89349。如果在發生率較低的國家進行篩檢（比如以台灣之吸菸者就2002年的發生率0.00171~0.00185而言），預期人年差異為0.0067~0.0071 年，而增加成本效用比（USD/QALY）為447258~752571。 結論：（1）台灣肺癌發生率就男性而言在公元2000年後已稍微下降，但就女性而言仍在稍微上升（2）相應於胸部愛克斯光較短的平均滯留時間，以胸部電腦斷層做肺癌篩檢相對有效（3）利用（1）及（2）的成果所進行的醫療經濟評估顯示以胸部電腦斷層做肺癌篩檢可增加預期人年差異及減少肺癌死亡率，但相較於其他癌症篩檢其增加成本效用比仍稍許偏高。
並列摘要
Background & Objectives: Lung cancer is the leading cause of cancer death world wide. There is no effective screening modality so far, and screening computed tomography (CT) is highly debated. There were two large ongoing randomized trials comparing CT v.s. observation (the Dutch-Belgian randomised lung cancer screening trial (NELSON)) or chest xray (CXR)(national lung screening trial, NLST), with results awaiting. The aims of this study were to (1) estimate the baseline incidence of lung cancer by age, period, cohort, (2) to estimate the mean sojourn time ( MST) of lung cancer by different screening modality; (3) to develop a decision modeling for economic evaluation of lung cancer screening by different intervals and screening modalities based on (1) and (2). Methods: We build up a five state Markov model to simulate the effectiveness (mainly presented as incremental life expectancy, ILE) and cost-utility analysis (presented as incremental cost-utility (in terms of quality-adjusted life year, QALY) ratio, ICUR) of lung cancer screening. Important parameters (incidence, mean sojourn time (MST), and treatment compliance) were established separately in order to implement the model. As to incidence, lung cancer cases (n=44139) diagnosed between 1996 and 2002 in Taiwan were analyzed by using a Bayesian age-period-cohort (BAPC) model. Age-adjusted standardized incidence rate (ASIR, per 100000 person-year) were calculated. As to MST by CT, data from six prospective CT screening studies were retrieved based on systematic literature review. The MST in association with the natural history of lung cancer depicted by a three-state Markov model was estimated with a Bayesian approach. As to MST by CXR, we collected data on demographic features, histology type, survival status, history of smoking, and asymptomatic or symptomatic status in light of chief complaint at diagnosis retrieved from medical records based on institutional cancer registry for lung cancer patients with prior non-diagnostic CXR (n=221) as an opportunistic screening cohort. The MST for the natural history of lung cancer underpinning a three-state Markov model was estimated with a Bayesian approach. As to treatment compliance, it was estimated from hospital cancer registry (from year 1991 to 2002, n=4565) via descriptive analysis. Other complimentary parameters were cited based on western literatures. Model validation was based on comparison between estimates and reported literatures for stage specific case identification, stage specific survival, and stage specific treatment related cost. Finally, effectiveness and cost-utility analysis of lung cancer screening were estimated via the above mentioned model and parameters for our primary and complementary scenarios. The primary scenario was a NELSON like (for selected smoker with median age 61 y/o) setting. Results: The prediction of our BAPC model was close to observation, and showed slightly decreasing incidence after around year 2000 for male but still slightly increasing for female. The median (95%CI) estimated annual incidence rate (/100000) of lung cancer in 61 year-old Taiwanese would be 102 (80~127) for male and 31 (22~44) for female in year 1996. The corresponding figures in year 2002 would be 123 (99~147) and 58 (45~75) respectively. The median (95% confidence interval, 95%CI) MST of lung cancer with screening CT was estimated as 2.06 (0.42~3.83) years. The median (95%CI) MST by screening CXR was estimated as 5.51 (4.04~7.12) months. Small cell lung carcinoma was even statistically significantly shorter MST than non-small cell lung carcinoma (3.01 months (3-3.98) versus 6.07 (4.44-8.25) months). The treatment compliance rates were 40% for advanced stage diseases. Model validation revealed comparable results in case finding (estimated/reported cases: 1.1%/1.3% for early stage disease and 1.4%/1.5% for all lung cancer), stage-specific survival (estimated/reported 5-year survival: (0.62~0.77)/(0.54~0.73) for early stage disease and (0~0.01)/(0.01~0.07) for advanced disease), and treatment related cost (US dollars, USD)(estimated/reported: 25183/25050 for early stage and 22372/20691 for advanced stage disease). In NELSON like setting, the median (95%CI) ILE would be 0.018 (-0.04~0.07) year. The mean ICUR (USD/QALY) would be 125171 in societal point of view (237145 if in payer’s point of view). In consider of the impact of different screening schedules, annual CT screening is associated with less lung cancer death when compared with three-yearly screening (percentage of lung cancer death: 2.1% vs 2.4%), but it is not cost-benefit when utility and cost was taken into account. In consider of different screening modality, CT screening vs CXR screening is more cost-benefit when compared with our main scenario (mean ICUR: 89348). If screening program was performed in a country with low incidence rate such as Taiwan (incidence of lung cancer for smoker in the NELSON like setting being 0.00171~ 0.00185 in year 2002), the median ILE would be 0.0067~0.0071 year. The mean ICUR (USD/QALY) would be 447258~752571. Conclusion: (1) Lung cancer incidence in Taiwan was slightly decreasing for male but slightly increasing for female. (2) The shorter mean sojourn time by using CXR as compared with MST by CT strongly suggests that CT screening may be more effective in early detection of lung cancer in screening. (3) Health economic evaluation based on (1) & (2) revealed that lung cancer screening with CT may lead to an increase in life expectancy and reduce lung cancer mortality, but the incremental cost-utility ratio remained high when compared with other cancer screening.